Enhanced hierarchical Bayesian modeling of mortality rates incorporating jumps: A multi-country perspective

The rapid aging of populations and global connectivity necessitates considering factors such as geographical proximity, socioeconomic conditions, and cultural connections in mortality forecasting. Additionally, catastrophes, pandemics, and political turmoil can disrupt the delicate balance of this complex system, introducing further layers of uncertainties to the already challenging task of predicting the future course of life expectancy. This project builds upon a hierarchical Bayesian random walk with drift (HBS-RW) model, inheriting its advantage of capturing the dependency structure of mortality rates across ages and populations. We further extend the model by adding jump components to account for extreme events in the data. From the numerical illustrations, the proposed model with jump components generally demonstrates superior performance compared to the HBS-RW model without a jump component and other Lee-Carter-based models in most conducted tests. Among the jump models, those incorporating trajectory jump effects yield the lowest average of mean absolute percentage error (AMAPE) overall. The choice between independent and dependent trajectory jump effects depends on the characteristics of the data. However, from a practical perspective, opting for the independent trajectory jump effect is favored, as it is significantly time efficient, compared to the models incorporating the dependent trajectory jump effect, in the maximum likelihood estimation for parameters.